Using evapotranspiration to assess drought sensitivity on a subfield scale with HRMET, a high resolution surface energy balance model

Highlights

• Develops HRMET, a remotely sensed surface energy balance model.

• HRMET estimates the energy balance at high resolution for homogeneous canopies.

• HRMET is applied to two cornfields in Wisconsin during the 2012 growing season.

• HRMET reveals persistent ET patterns throughout growing season.

• ET patterns are used to map subfield scale sensitivity to drought.

Abstract

Evapotranspiration (ET) rates provide a valuable within-season indicator of plant productivity, as well as data on fluxes of water in a landscape. Applying remote sensing for ET estimation has potential to improve the sustainable management of water resources in agricultural settings. Most current ET models, however, rely on ‘dry’ and ‘wet’ pixels within a given scene to partition turbulent fluxes between latent and sensible heat, thus limiting their ability to map ET throughout the growing season at extremely high (meter scale) spatial resolutions. Here, we develop a field-validated surface energy balance model, High Resolution Mapping of EvapoTranspiration (HRMET), which requires only basic meteorological data, spatial surface temperature and canopy structure data. We use HRMET to estimate ET rates over two commercial cornfields in south-central Wisconsin during the 2012 growing season, which was characterized by severe drought. HRMET results indicate that the magnitude of within-field variability in ET rates was primarily driven by water availability. The application of remotely sensed data to precision agriculture has also been hampered by turnaround time between image acquisition and availability. We introduce relative ET (ET_R), which enables comparison of ET rates between image dates by normalizing for variability caused by weather and crop stage. ET_R also provides an intuitive, index-like metric for evaluating spatial variability in ET on a subfield scale. ET_R maps illuminate persistent patterns in ET across measurement dates that may be driven by soil water availability and topography. ET_R is used to develop a novel paired-image technique that can map subfield sensitivity classes to stressors such as drought. Sensitivity class mapping can be used to circumvent issues related to turnaround time to facilitate the incorporation of remotely sensed data into precision agriculture.

Introduction

Spatially distributed estimates of evapotranspiration (ET) are a critical agricultural resource (Gowda et al., 2007). Farmers, agronomists, and water resource managers need accurate data about plant water use to evaluate irrigation requirements, respond to drought or precipitation events, and forecast year-end yields (Hiler et al., 1974, Idso et al., 1977, Kanwar et al., 1984, Akhtar et al., 2013, Kresovic et al., 2013). ET estimates can also enhance ecosystem and agroecosystem models by providing estimates of mass and energy fluxes over the landscape (Kucharik and Brye, 2003, Kucharik and Twine, 2007, Soylu et al., 2014).

Spatially distributed estimates of ET are particularly useful for applications in precision agriculture, for which high spatial resolution is a necessity. The fundamental theory behind precision agriculture is that the optimum inputs vary spatially on the subfield scale, depending on factors such as soil texture, soil organic matter, water availability, topography, and more (Wibawa et al., 1993, Kravchenko and Bullock, 2000, Schmidt et al., 2002, Kravchenko et al., 2003, Schepers et al., 2004). Many techniques have been used to estimate spatially distributed ET rates from both agricultural and natural settings using remotely sensed data. Glenn et al. (2007), Gowda et al. (2007) and Maes and Steppe (2012) provide reviews of the techniques available. One class of ET models are surface energy balance models, which attempt to estimate the partitioning of available energy (A) at the land surface to determine the relative magnitudes of the sensible heat flux (H) and latent heat flux (λET), which is directly related to the ET rate. The general surface energy balance equation is as follows:

$$R-G=A=H+\lambda ET$$

where R is equal to net incoming radiation, G is the heat flux into the subsurface, and all units are in [W m⁻²].

Surface energy balance models relying on remotely sensed data typically require surface reflectance data in one or more wavelengths and surface temperature data, all of which are available on common sensor platforms. Commonly used models include SEBAL (Bastiaanssen et al., 1998), METRIC (Allen et al., 2007), and SEBI (Menenti and Choudhury, 1993, Roerink et al., 2000). These models rely on the presence of a ‘dry’ (also referred to as ‘hot’) and ‘wet’ (‘cool’) pixel within the image to define the extremes of surface temperature, where all available energy is apportioned to sensible heat or latent heat fluxes, respectively. Other energy balance models, such as ALARM (Suleiman & Crago, 2004), have been shown to work in the absence of dry and wet pixels, but assume a closed canopy to simplify canopy transport physics. Previous energy balance models have been applied effectively over large areas (Elhaddad et al., 2010, Bastiaanssen et al., 2012) and small spatial scales with diverse land surface cover (Loheide and Gorelick, 2005).

However, these widely used energy balance techniques are typically applied at regional scales using moderate spatial resolution (30–1000 m) satellite data, and are structurally incompatible with high-resolution remote sensing of crops. Though field-scale management based on ET estimates have been practiced with some success (Folhes et al., 2009, Ko and Piccinni, 2009; M. C. Anderson et al., 2012a; R. G. Anderson et al., 2012b), subfield-scale precision management based on remotely sensed ET estimates has not yet been attempted. Many agricultural landscapes are fairly homogeneous, making it difficult or impossible to identify dry and wet pixels within the same image–for high resolution image collection required for precision agricultural management, image extent can be limited to ≤1 km². For most agricultural settings, completely closed canopies are rarely available during the early growing season, and dry or fallow open canopies are rarely available late in the growing season, limiting these models’ application to transient dynamics in ET. Furthermore, while satellite-based estimates of ET are useful for landscape scale studies, even the highest-resolution multispectral and thermal satellites are insufficiently detailed for most subfield scale applications.

To address these issues unique to ET estimation at small spatial scales with homogeneous canopies, we developed a mixed-input surface energy balance model for the High Resolution Mapping of EvapoTranspiration (HRMET). HRMET employs high-resolution (∼1–2.5 m) surface temperature data collected at a subfield scale in conjunction with 30 m resolution Landsat imagery and local meteorological data to estimate the surface energy balance. HRMET uses a combination of physical and empirical relationships to calculate A, and an iterative approach to calculate H which does not require wet and dry pixels. λET is estimated as the residual and converted to an instantaneous ET rate. A full model description is found in Section 3.

While spatially distributed estimates of ET are an essential tool for hydrologists and water resource managers (M. C. Anderson et al., 2012a), a major limitation of applying ET mapping to agricultural management at all scales is the difficulty in making comparisons between image collection dates. A “good” ET rate can vary widely between dates due both to stress-related changes (e.g. stomatal closure caused by drought) and changes out of a farmer's control (e.g. crop growth stage, weather conditions). This makes management based on ET data from different measurement dates challenging and non-intuitive.

For this reason, most remotely sensed analyses of crop drought response use stress indices such as the crop water stress index (CWSI; Jackson et al., 1981), water deficit index (Moran et al., 1994), perpendicular drought index (Shahabfar and Eitzinger, 2011), or the integrated surface drought index (Wu et al., 2013). The primary advantage of these indices is a low data requirement and intuitive interpretation; the CWSI, for example, ranges from 0 to 1. To bridge this gap between the utility of physically-based ET estimates valued by hydrologists and intuitive stress indices valued by land managers, we introduce relative ET (ET_R). ET_R is a novel way to interpret HRMET's spatially distributed ET estimates and allows us to easily compare measurement dates in the manner of a crop stress index. We compare corn response to a severe drought during the 2012 growing season, using a study site in south-central Wisconsin as an exemplar. ET_R allows us to identify persistent patterns of ET on a subfield scale, which could be used to target areas of the field for precision management such as variable rate water application during drought (King and Kincaid, 2004). We also demonstrate how deviations from these patterns can be used to identify drought-sensitive and moisture-sensitive areas on a subfield scale. Identification of these stress-sensitive areas can guide precision management decisions, even in the absence of real-time remotely sensed data.